The Glucose Management Indicator: Time to Change Course?

Continuous glucose monitoring (CGM) technology has been revolutionary for patients with type 1 diabetes and their families, and there is emerging evidence that CGM can be useful in some patients with type 2 diabetes (1). Randomized clinical trials have demonstrated that CGM use can improve glycemic control, reduce episodes of hypoglycemia, and improve the quality of life of people living with diabetes who are on intensive insulin regimens (2–12). CGM technology has improved substantially in ease of use and accuracy over the past two decades. The latest generation of devices, which do not require finger-stick calibration and have been integrated with insulin pumps, have been a major advance for patients and the field. Nonetheless, this technology remains relatively new, and there is debate regarding the optimal use and interpretation of the different metrics calculated from CGM data.

CGM systems generate large amounts of data, and interpreting this detailed information on glucose patterns can be complex for patients and providers. The glucose management indicator (GMI) is calculated from an equation that translates CGM mean glucose into an estimation of the hemoglobin A_1c (HbA_1c) value (13). Laboratory measurement of HbA_1c is the standard approach to monitoring glycemic control in people with diabetes, and GMI was initially proposed to help simplify and aid in the interpretation of CGM data.

This Perspective discusses the origins of GMI, interrogates its strengths and limitations, and lays out a case for rethinking the use of GMI when using CGM to optimize glucose control and prevent complications in people with diabetes. It is shown here that it is preferable to use CGM mean glucose rather than converting this glucose to an estimated HbA_1c (eA_1c) or GMI. Using mean glucose as a complement of laboratory HbA_1c testing in patients who use CGM systems can reduce patient and provider confusion and should improve glycemic management.

Laboratory HbA_1c is central to the care of patients with diabetes. For over three decades, HbA_1c has been the standard test used to monitor glycemic control. Randomized clinical trials have demonstrated that lowering HbA_1c reduces the risk of major complications (14–19). HbA_1c assays are standardized and traceable to a reference method, and proficiency testing is conducted to monitor the effectiveness of standardization (20). Using this approach to standardization, HbA_1c can be measured with tremendous accuracy and precision (modern assays have coefficients of variation <3%).

HbA_1c is a measure of chronic hyperglycemia, reflecting the nonenzymatic glycation of hemoglobin in the red blood cells over the past 2–3 months. It is strongly associated with future microvascular and macrovascular disease, potentially because it reflects protein glycation, which plays a role in the development of diabetes complications (21). While the major determinant of elevated HbA_1c is indisputably exposure to high concentrations of blood glucose, HbA_1c is an indirect measure of hyperglycemia. Nonglycemic factors can influence HbA_1c, including red blood cell turnover, other red blood cell characteristics, and genetic variation in hemoglobin. These nonglycemic factors can affect the association of HbA_1c with “true” average glucose exposure, particularly in the low (nondiabetic) range (22).

In 2008, the ADAG (A1C-Derived Average Glucose) study rigorously quantified the association of mean glucose with HbA_1c (23). Mean glucose in the ADAG study was calculated from a combination of measurements from an early CGM system and capillary glucose (finger stick) collected in 507 adults (84% were adults with type 1 or type 2 diabetes and 16% were adults without diabetes; 83% were White adults), intermittently, during a 3-month period. This older system required calibration several times a day using a self-monitoring glucose meter. The original intent of the ADAG study was to provide estimates of mean glucose based on HbA_1c; this contrasts with the GMI equation, which estimated HbA_1c as a function of mean glucose. Thus, the linear regression equation from the ADAG study is regressed in the direction opposite that of GMI. (Note that least-squares linear regression is not symmetric, that is, regressing A on B will produce a different regression line than regressing B on A.) Estimated mean glucose values from the ADAG equation are provided in the American Diabetes Association (ADA) Standards of Care in Diabetes and offer a way for patients and providers to translate laboratory HbA_1c to mean glucose.

Prior to the ADAG study, estimates of average glucose corresponding to different values of HbA_1c were provided in pre-2009 versions of the ADA Standards of Care and were derived from participants with type 1 diabetes in the Diabetes Control and Complications Trial (DCCT). DCCT investigators evaluated the correlation of mean glucose values derived from 7-point capillary glucose profiles and laboratory HbA_1c. Like the ADAG equation, the DCCT equation regressed mean glucose on HbA_1c and established a regression equation to translate HbA_1c into average glucose equivalent values (24).

DCCT and ADAG were conducted during a time when obtaining many glucose measurements, throughout the day and night and over long periods of time sufficient to calculate a typical or average glucose in an individual, was difficult and burdensome. Modern CGM sensors are simple to use, provide an interstitial glucose reading every minute or every few minutes for up to 14 days, and do not require finger-stick calibration. Thus, it makes sense to update our understanding of the mathematical association of average glucose with HbA_1c using modern CGM technology.

In 2017, a new equation was proposed for translating mean CGM glucose into estimated HbA_1c, or eA_1c (25). This 2017 equation estimated HbA_1c from CGM mean glucose derived exclusively from three randomized trials of the Dexcom G4 CGM system in adults with type 1 diabetes. Subsequently, this new equation was incorporated into standard CGM or average glucose profile (AGP) reports (26). AGP reports provide visual and statistical summaries of the data from CGM systems, and early AGP reports included eA_1c values based on the 2017 equation.

The 2017 equation was, however, renamed in 2018 in response to concerns raised by the U.S. Food and Drug Administration (FDA) that reporting CGM mean glucose to patients as eA_1c could be confusing (13). To address FDA concerns, it was proposed that estimates of HbA_1c from a new 2018 equation be called the GMI. Data from an additional Dexcom G4 CGM trial (12) were incorporated into the 2018 equation at that time.

The GMI metric is now used by all major CGM manufacturers and printed on AGP displays for patients. Because mean glucose from CGM is not recognized as a typical glycemic control target, an advantage of GMI is that it translates mean glucose values to an estimated HbA_1c. The HbA_1c value is familiar to patients and providers and is a validated surrogate end point for clinical trials of glucose-lowering interventions. HbA_1c targets are well established, and the GMI is used as a substitute for monitoring glycemic control when laboratory HbA_1c is not available. The 2023 ADA guidelines for glycemic assessment recommend, “Assess glycemic status (A1C or other glycemic measurement such as time in range or glucose management indicator) at least two times a year in patients who are meeting goals . . . and at least quarterly and as needed in patients whose therapy has recently changed and/or who are not meeting glycemic goals” (Recommendations 6.1 and 6.2 in ElSayed et al. [27]).

It is worth noting that while GMI is printed in all AGP reports, it is based exclusively on data from Dexcom CGM trials. GMI is based on a linear regression equation from a study population of just over 500 adult patients with type 1 or type 2 diabetes who participated in four different trials that examined the effectiveness of the Dexcom G4 CGM (8,9,12,28). CGM mean glucose was compared with HbA_1c measurements obtained after ∼48 days (range, 13–89 days). The average age of participants in these trials ranged from ∼45 to 60 years, and >80% of participants in the three U.S. trials were White adults; the one trial of German patients did not report race or ethnicity.

Estimated HbA_1c based on mean glucose from the ADAG, DCCT, and GMI formulae are generally similar, especially at mean glucose values <200 mg/dL. For a mean glucose of 180 mg/dL (10.0 mmol/L), the eA_1c using the DCCT equation is 7.2%, for ADAG is 7.9%, and for GMI is 7.6% (Table 1 and Fig. 1). This suggests that the mathematical relationship between mean glucose and eA_1c is reasonably robust across these three study populations, which were made up predominately of adults with type 1 diabetes. Nonetheless, the information needed to generate CIs and understand the precision of these estimates is not available. Further, recall that for ADAG and DCCT, the original equations are regressed in the opposite direction, which is one reason they do not fully align with GMI.

Since GMI is being used in individual patient reports and as a substitute for laboratory HbA_1c test results in some settings, an important question is how accurate is GMI as an estimate of HbA_1c in individual patients. Prediction for an individual patient is inherently more difficult than quantifying population associations. In the 2018 article by Bergenstal et al. (13) on GMI, no estimates of precision or model fit are available, although the authors noted that 28% of their study participants had clinically significant discordance between GMI and laboratory HbA_1c, defined as a difference between these two measures of ≥0.5 percentage points (%-points). The discussion below puts this 28% discordance into context and addresses the performance of GMI in external populations.

Since its introduction in 2018, there have many efforts (news articles, websites, and videos) to help educate patients and providers about the GMI and its interpretation. There have also now been many studies evaluating the relationship of GMI to HbA_1c in different study populations, using different CGM sensors and including a broad age range of patients with and without diabetes.

I identified close to two dozen studies evaluating the concordance of GMI to measured HbA_1c in different populations (Table 2). Using CGM sensors from different manufacturers that were conducted in populations of adults and children with type 1 diabetes, people with type 2 diabetes not using insulin, people with comorbidities such as kidney disease, and people without diabetes, these studies have consistently shown a high percentage of individuals with a clinically significant difference (≥0.5 %-points) between GMI and HbA_1c, with estimates ranging from 26% to 68% (Fig. 2). These studies suggest that there is frequent discordance between HbA_1c and GMI.

To further understand this discordance and potential variability in GMI, I evaluated how GMI might differ in comparisons of two different CGM sensors. In the absence of systematic or random error (assuming complete accuracy of CGM technology), GMI from two different CGM sensors should be identical, since both are calculated from a mathematical transformation of mean CGM glucose. However, we know that CGM technology is not perfect. By comparing CGM glucose or GMI from two different sensors worn at the same time, it is possible to quantify the degree of random and systematic error one might expect is inherent to this technology.

To address this question, data from 176 participants in the Hyperglycemic Profiles in Obstructive Sleep Apnea (HYPNOS) trial were analyzed (29–31). The HYPNOS trial included participants aged 30–70 years (mean age 60 years) with type 2 diabetes who were not taking insulin. These participants wore two different CGM sensors (Abbott Libre and Dexcom G4) at the same time (only simultaneous periods were included in this analysis). Deming regression was used to characterize the associations of GMI from the two different sensors (32,33). In contrast to ordinary least-squares regression, Deming regression is symmetric and assumes that both variables being compared are measured with error (whereas least-squares regression is asymmetric and assumes only the outcome is measured with error). A Bland-Altman plot to show the differences between the two GMI estimates was also used.

During up to 4 weeks of wear time, the mean GMI for the Libre sensor was 7.04% (SD 1.00%). The mean GMI for the Dexcom sensor was 7.06% (SD 0.89%). While the means were similar, there was substantial scatter around the regression line (Fig. 3A), indicating that GMI-Abbott and GMI-Dexcom frequently do not give the same result, even when calculated from CGM values obtained during the same period in the same person. A good way to quantify this error is using the root mean squared error (RMSE) and the SD of the differences. The RMSE indicates the typical difference, or deviation, from the regression line. Here, the RMSE was 0.27 %-points. In other words, GMI calculated from CGM mean glucose from these two sensors when worn on the same person at the same time will typically disagree by ∼0.3 %-points. Similarly, the SD of the differences was 0.4 %-points. Overall, 26% (46 out of 176) of HYPNOS participants had differences of ≥0.5 %-points when comparing GMI from the two different CGM sensors (Fig. 3B).

Prior studies have compared GMI to laboratory HbA_1c. Here, we compared GMI-Dexcom and GMI-Abbott and showed that GMI from different sensors worn on the same person at the same time frequently do not align. Thus, these data also demonstrate that when HbA_1c and GMI differ, variability and inaccuracy in CGM glucose may be a source of the problem.

The discordance between GMI and measured HbA_1c is often attributed to inaccuracy in HbA_1c, but it is important to note that international standards for ensuring accuracy of laboratory HbA_1c measurements are well established (34). However, in any given individual, red blood cell turnover and other nonglycemic determinants of HbA_1c may differ, potentially contributing to discrepancies between mean glucose and HbA_1c.

In contrast to laboratory HbA_1c, standards for determining the accuracy of CGM sensors are less well established and vary across countries, traceable reference methods are not available, CGM sensor algorithms and approaches to calibration are manufacturer specific, and there are limited studies with head-to-head comparisons of different CGM sensors (35–37). The few studies evaluating within- and between-sensor differences have shown significant variability (38–40). Discordance may also be partially explained by FDA requirements for CGM sensor accuracy, which generally specify that CGM glucose readings must be within 15–20% of reference glucose (venous or capillary glucose), although the specific requirements differ depending on the glucose concentration (41). Other sources of error may include the lag time between interstitial glucose and venous glucose, which may vary between individuals. Finally, it is worth noting that venous glucose and capillary glucose can have clinically significant differences and high interindividual variability, which can affect estimates of CGM accuracy (42).

Underlying reasons for within- and between-sensor variability in CGM glucose reflect limitations of the technology, lack of standardization of interstitial glucose methods, approaches to calibration, and differences in the proprietary algorithms that generate glucose readings (43). Patient-level factors can influence the accuracy of CGM systems, including bouts of physical activity and other situations that result in rapid changes in glucose, blood flow, and local conditions (44) (Table 3). Other contributors to discrepancies between HbA_1c and CGM glucose or GMI include the duration of CGM wear and timing of the HbA_1c measurement relative to the period of CGM wear.

Discordance between HbA_1c and GMI is commonplace and can be substantial. This creates confusion for patients and providers. Where should we go from here? We could improve GMI. Studies using the latest generation of sensors from all the different manufacturers would be required. Clinical equations would need to be developed and then validated in large, diverse populations. Prior studies have incorporated information on red blood cell physiology and life span to improve the match of CGM mean glucose to HbA_1c (22,45,46). Other studies have proposed population-specific GMI equations (47,48), but is this worthwhile? Even if we improve GMI, we will always see some degree of discordance between CGM mean glucose and HbA_1c. HbA_1c and CGM glucose are distinct entities with different sources of variability.

A fundamental issue is that GMI may not solve the concerns raised by the FDA regarding eA_1c. Putting mean glucose on the same scale (units) as HbA_1c can create the impression that GMI and HbA_1c are equivalent and interchangeable. Patients might therefore cancel HbA_1c testing. Confusion about treatment targets may arise: should the patient use the laboratory HbA_1c or the GMI? There can also be confusion about what to do when there is discordance between measured HbA_1c and GMI, a common scenario (Fig. 2). Finally, the GMI equation assumes that it performs equally well across CGM systems and that the association between CGM glucose and HbA_1c is the same for all people, including children and adolescents, although no pediatric patients were included in the studies used to derive GMI.

An alternative to relying on GMI is to acknowledge that HbA_1c and mean glucose are different entities with their own strengths and limitations. Compared with CGM systems, which can cost several hundred dollars per month, laboratory HbA_1c is inexpensive (typically about $10/test) and widely available to patients with diabetes. HbA_1c is measured with precision, is well standardized, and is strongly linked to clinical outcomes. Major benefits of CGM use for patients are the detailed and nuanced data on glucose patterns, information on trends, and real-time, continuous feedback. Many CGM sensors are customizable, wirelessly transmit data to smart phones or other devices, and provide alerts for glucose highs and lows. In patients using CGM sensors, health care professionals should emphasize the importance of monitoring mean glucose, time in range, real-time trends in glucose, and other CGM metrics, including a holistic evaluation of overall glucose patterns.

Patients and providers should be aware that mean glucose will typically track with HbA_1c, but this is not always the case. Discordance can arise from patient-related factors, CGM-related factors, and/or HbA_1c-related factors (Table 2). HbA_1c and CGM mean glucose are correlated, but these correlations are imperfect. This imperfect correlation also suggests we cannot just assume that CGM metrics will be as strongly predictive of complications as HbA_1c. CGM metrics are not yet linked to long-term clinical outcomes in rigorous, prospective studies. Long-term studies of different CGM metrics and clinical complications, with head-to-head comparisons to HbA_1c, are needed.

For GMI, it may be time to change course. There is now a large body of literature demonstrating that GMI and HbA_1c are frequently highly discordant. GMI does not reflect protein glycation and does not perform well as a substitute for HbA_1c. Instead of relying on GMI, we should establish CGM mean glucose ranges to use as treatment targets. For patients who will benefit from CGM, especially those at risk for hypoglycemia, we should embrace mean CGM glucose, time in range, and other metrics.

ADA guidelines currently provide a table based on data from the ADAG study that shows how a given HbA_1c is likely to relate to mean glucose ranges (27). However, showing ranges of CGM mean glucose for typical ranges of laboratory HbA_1c is more helpful, since both measures have error. We should continue to provide in the guidelines side-by-side estimates of mean glucose and HbA_1c, but we should update these with data from studies using modern HbA_1c assays and the latest generation of CGM sensors from different manufacturers. To enhance the accuracy and generalizability of these estimates, we should rely on data collected in a standardized fashion from diverse populations (i.e., people with different forms of diabetes, people not using insulin, children, people from diverse socioeconomic backgrounds, and people in underrepresented minority populations). Finally, we need to improve clinical guidance on what to do when CGM glucose and HbA_1c do not line up, acknowledging that both entities have underlying sources of error.

If we are to embrace CGM for broader populations of patients with diabetes, including those not using insulin, we need studies that document the cost-effectiveness of CGM use, demonstrate improvements in the health of patients who are at low risk of hypoglycemia, and rigorously link CGM metrics to long-term complications with head-to-head comparisons to HbA_1c. The optimal use of CGM and its metrics in the care of patients with diabetes requires knowledge of both the strengths and limitations of this groundbreaking technology.

The interrogation of the strengths and limitations of GMI in this Perspective suggests that GMI is a concept that has outlasted its usefulness. Focusing on CGM glucose rather than GMI is not as big a change as it might seem. GMI is a linear transformation of mean glucose. Thus, when we are relying on GMI to make decisions, we are already relying on mean glucose to guide care. Leaving mean glucose in its raw form is simple and reinforces that glucose and HbA_1c are distinct. Instead of replacing HbA_1c with CGM metrics, as some experts have recommended (49), we need to ensure that providers and patients are using HbA_1c and CGM together, in a complementary manner, to optimize glucose control and prevent complications in people with diabetes.

Acknowledgments. The author thanks Dan Wang for her assistance with the statistical analyses, Henry Zhao for his help in compiling data from studies evaluating GMI and HbA_1c discordance, and Dr. Michael Fang for his helpful comments on an early draft.

E.S. is an editor of Diabetes Care but was not involved in any of the decisions regarding review of this manuscript or its acceptance.

Funding. E.S. is supported by National Institutes of Health (NIH), National Institute of Diabetes and Digestive and Kidneys Diseases, grants R01 DK128837 and R01 DK128900; NIH, National Heart, Lung, and Blood Institute, grants K24 HL152440 and R01 HL158022; and NIH, National Institute on Aging, grant RF1 AG074044. Abbott Diabetes Care provided CGM systems and self-monitoring blood glucose supplies for the investigator-initiated research in the HYPNOS trial. Dexcom provided continuous glucose monitoring systems in the HYPNOS trial at a discount.

Duality of Interest. No potential conflicts of interest relevant to this article were reported.

Handling Editors. The journal editor responsible for overseeing the review of the manuscript was Matthew C. Riddle.